LIVE
Google Introduces "Gemini 4 Carbon" as Next Step in Its Model Lineup28/10/26 · Google|DeepMind unveils Genie 4, a new generation of world model27/10/26 · Google DeepMind|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|Robot policy trained only in simulation matches top real-world systems26/10/26|OpenAI unveils GPT-6.1 Terra, a model built for computer use26/10/26 · OpenAI|Google's AI gives evasive answers about who won the 2020 US election28/10/26 · Google|NVIDIA introduces Nemotron 3.5 Transcribe, a promptable multilingual transcription model28/10/26 · NVIDIA|OpenAI and ServiceNow deepen partnership on enterprise workflows28/10/26 · OpenAI|Failed startups' data finds a second life training AI models28/10/26|Anthropic expands AI for Science program to 20 foundations and research institutions28/10/26 · Anthropic|Anthropic shares early results on using Claude to accelerate drug development27/10/26 · Anthropic|GEPA-Curator: Filtering Failures to Better Train LLM Agents27/10/26|Google Introduces "Gemini 4 Carbon" as Next Step in Its Model Lineup28/10/26 · Google|DeepMind unveils Genie 4, a new generation of world model27/10/26 · Google DeepMind|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|Robot policy trained only in simulation matches top real-world systems26/10/26|OpenAI unveils GPT-6.1 Terra, a model built for computer use26/10/26 · OpenAI|Google's AI gives evasive answers about who won the 2020 US election28/10/26 · Google|NVIDIA introduces Nemotron 3.5 Transcribe, a promptable multilingual transcription model28/10/26 · NVIDIA|OpenAI and ServiceNow deepen partnership on enterprise workflows28/10/26 · OpenAI|Failed startups' data finds a second life training AI models28/10/26|Anthropic expands AI for Science program to 20 foundations and research institutions28/10/26 · Anthropic|Anthropic shares early results on using Claude to accelerate drug development27/10/26 · Anthropic|GEPA-Curator: Filtering Failures to Better Train LLM Agents27/10/26|
Toolofficial publisher

Task, sous-agents (Claude Code)

Tool : A precise action the AI can call · requires claude-code

Launches an autonomous sub-agent with its own tools.

Powerful for parallelizing work; it inherits permissions from the parent, which introduces risk.

Docs
Overall score
6.9 /10
Security: one caveat, see the card
Utility
8.0
Reliability
6.0
Security
One caveat
Ease
5 min · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

When a side task risks flooding the main conversation with logs, search results, or file contents you will not reference again, Claude Code subagents handle that work in their own context window and return only a summary.

Each subagent runs with its own system prompt, restricted tool access, and independent permissions. Claude automatically decides to delegate a task to a subagent by reading its description, then returns only the result to the main conversation.

This tool is aimed at developers working on complex codebases who want to parallelize exploration, planning, and execution without saturating their main context. It also helps control costs by routing certain tasks to cheaper models such as Haiku.

However, subagents are not suited for fully independent sessions that need to communicate with each other or be monitored from a dashboard: other Claude Code features (background agents, agent teams) address those needs.

What it offers

  • ·Built-in subagents (Explore, Plan, General-purpose) activated automatically based on task type.
  • ·Custom subagent creation via a Markdown file with YAML frontmatter, stored in .claude/agents/ or ~/.claude/agents/.
  • ·Tool access restriction per subagent (for example, read-only for Explore and Plan).
  • ·Per-subagent model selection to control costs (Haiku, Sonnet, Opus).
  • ·Configurable scope: subagent available for a single project, all your projects, or an entire organization via managed settings.
  • ·Temporary subagent definition via the --agents CLI flag, without writing to disk.
  • ·Startup warning when combined descriptions of custom subagents exceed 15,000 tokens.

In practice

Price
Free
Commercial use
Commercial use: yes
Available on
claude-code
Host
claude-code
FR interface
No

Security and data

Rating
Security: one caveat, see the card
Capabilities
readswritesexecutesnetwork
Your data
Data stays local
Maintenance
Active

Links and repository checked on 25 October 2026 · added on 1 July 2025

How to install Task, sous-agents (Claude Code)

1. Make sure Claude Code is installed and up to date (v2.1.198 minimum recommended). 2. To create a subagent, ask Claude: "Create a [description] subagent in ~/.claude/agents/" or write a Markdown file with YAML frontmatter yourself. 3. Review the generated file at ~/.claude/agents/subagent-name.md and confirm the name, description, tools, and model fields. 4. To limit risk, restrict tools to the strict minimum (for example tools: Read, Grep, Glob for a read-only agent). 5. Restart Claude Code if the ~/.claude/agents/ directory did not exist before the current session. 6. Test by asking Claude to delegate a task to the new subagent and verify that the delegation appears in the transcript.
.claude/agents/*.md

Official documentation ↗

In the same category

Whole category Orchestrate →